Thank you for starting this discussion, Seonggon. I am also +1 (non-binding) for option 1: pluggable IMetaStoreClient. This is the most realistic path for actual adoption, considering multiple existing integrations. (Disclaimer: I work for a company that has implemented a variant of HIVE-12679 in a private fork.)
Chris Nauroth On Sun, May 4, 2025 at 12:19 AM Shohei Okumiya <oku...@apache.org> wrote: > Hi Seonggon, > > Thanks for being involved with the ticket, trying a PoC, and > initiating the discussion on ML. > > I am currently biased toward Option 1 for some reasons. > - Option 1 likely has some existing users > - Option 1 can improve the maintainability of > SessionHiveMetaStoreClient and its family. It's nice, even if we find > Option 2 is more convenient later > - The primary disadvantage of Option 1, i.e., many overridden methods > in IMetaStoreClient, can be mitigated if we gradually deprecate and > clean up these methods > - In the worst case, we can additionally implement Option 2. We can do > what we want to do, and some tech debts will be gone > > Best, > Okumin > > On Wed, Apr 30, 2025 at 8:39 PM Denys Kuzmenko <dkuzme...@apache.org> > wrote: > > > > +1 for Option 1: Make IMetaStoreClient pluggable >
